About HarnessHub

The directory of agentic harness engineering

HarnessHub is a curated directory of the agentic harness layer — the engineering discipline of wrapping large language models in the scaffolding that makes them reliably useful: orchestration loops, tools, memory, sandboxes, protocols and evaluation.

The field is moving fast and scattered across thousands of repositories, startups, meetups and announcements. HarnessHub gathers the people and projects shaping it into one searchable place, so you can answer questions like: which orchestration frameworks are actually maintained? Who is building memory? Which events are worth flying to? What shipped this quarter?

What we track

• StartupsCompanies building the harness layer — 69 tracked, from orchestration platforms to evals and sandbox providers.
• Open-source projectsFrameworks, protocols, memory systems, agents and evals tooling — 114 repositories with stars, licenses and status.
• EventsConferences, meetups and hackathons where harness engineering happens — 22 tracked, with dates and locations.
• NewsNotable launches, fundings and protocol milestones that shaped the stack — 30 items.
• MapThe whole ecosystem plotted geographically — where harness companies actually cluster.

How entries are organized

Every startup and project is tagged across the layers of the harness stack — an entry can (and usually does) belong to more than one:

The harness landscape page walks through each layer and who is building it.

How it is put together

HarnessHub is curated by hand. Entries are researched from primary sources — company sites, GitHub repositories, official event listings — and reviewed periodically. GitHub star counts are refreshed from the GitHub API; funding and valuation figures are approximate public reporting, not audited numbers, so treat them as orientation rather than fact.

Inclusion is editorial: we list companies and projects whose primary business is the harness layer (not end-user AI apps in general), events with substantive agent-engineering content, and news that changed the shape of the stack. Suggestions and corrections are welcome — the data lives in a handful of typed files and is designed to be extended.